benchmark.rb

require 'benchmark'

def benchmark(setting, &block)
  data = (1..10_000_000).map { block.call }

  Benchmark.bm(15) do |x|
    x.report("compact #{setting}") do
      data.dup.each do |d|
        [d[0],d[1]].compact.min
      end
    end

    x.report("ternary #{setting}") do
      data.dup.each do |d|
        d[0] ? [d[0],d[1]].min : d[1]
      end
    end
  end
end

benchmark('exists'){ [rand,rand] }
benchmark('nil'){ [nil,rand] }

results.txt

                     user     system      total        real
compact exists   8.620000   0.690000   9.310000 ( 12.577204)
ternary exists   7.410000   0.230000   7.640000 (  7.646824)
                     user     system      total        real
compact nil      4.450000   0.020000   4.470000 (  4.473185)
ternary nil      1.290000   0.000000   1.290000 (  1.291536)

I expanded the sample size from 1,000 to 10,000,000, because comparing microseconds is just silly to me. With larger sample sizes we see a roughly 1.5:1 and 3.5:1 difference between compact and ternary.